Over the past 18 years online advertising has grown to a $70 billion industry worldwide annually. Despite this impressive growth, online advertising faces many (and some would say traditional) challenges including how to measure the efficiency or the potential loss of sales caused by the inefficient use of advertising dollars. Consequently, it is vital to measure, maximize, and benchmark the efficiency of advertising media expenditures. This tutorial introduces the field of econometrics as a means of measuring the effectiveness of digital marketing. Econometrics is a field that extends and applies statistical methods to the analysis of economic phenomena. In that vein, econometrics goes beyond traditional statistics and explicitly recognizes the complexities of human behavior. Consider for example the impact of deep discounts on survival of restaurants. Struggling businesses are more likely to offer these deep discounts and eventually fail. A naïve application of statistical techniques will overestimate the impact of deep discounts on business survival. In this case, the discounts are an endogenous variable as compared to an exogenous variable. This type of specification error highlights why we need a deeper look at the variables that go into statistical models. Econometrics addresses these and other issues in a formal and rigorous manner.

Local search engines allow users to search for entities such as businesses in a particular geographic location. To improve the geographic relevance of search, user feedback data such as logged click locations are traditionally used. In this paper, we use anonymized mobile call log data as an alternate source of data and investigate its relevance to local search. Such data consists of records of anonymized mobile calls made to local businesses along with the locations of celltowers that handled the calls. We model the probability of calls made to particular categories of businesses as a function of distance, using a generalized linear model framework. We provide a detailed comparison between a click log and a mobile call log, showing its relevance to local search. We describe our probabilistic models and apply them to anonymized mobile call logs for New York City and Los Angeles restaurants.